That's really the issue, isn't it? That nearly all data is at least somewhat noisy. Making inferences from it is just a big blur without a systematic way to handle it.